huiyu8794 / ldcnet

Learnable Descriptive Convolutional Network for Face Anti-Spoofing (BMVC'22)

LDCNet

Learnable Descriptive Convolutional Network for Face Anti-Spoofing (BMVC'22)

Feature maps extracted by LDC

Generate by Low_level_features_visualization.py

Learnable Descriptive Convolution

Network Architecture

Requirements

grad_cam==1.3.5
matplotlib==3.5.2
numpy==1.22.3
scikit_learn==1.1.2
torch==1.12.0
torchvision==0.13.0

Training

Step 1: run Amap_train.py to get pretrained model for producing activation map

Step 2: run train.py to train LDCNet

Testing

run test.py

More visualization: Feature maps extracted by LDC

Live images:

Print images (grid artifacts):

Replay images (moiré patterns):

Citation

If you use the LDC, please cite the paper:

@article{huang2022learnable,
  title={Learnable Descriptive Convolutional Network for Face Anti-Spoofing},
  author={Huang, Pei-Kai and Ni, Hui-Yu and Ni, Yan-Qin and Hsu, Chiou-Ting},
  year={2022}
}

Contact us

We are students from MPLAB at National Tsing Hua University.
Huang, Pei-Kai [email protected]
Ni, Hui-Yu [email protected]
Chong, Jun-Xiong [email protected]

ldcnet's Issues

Pretrained LDCNet

Hi,

I see the pretrained model you provide is for visualization, could you provide the weights of your LDCNet model?

Thank you